1. Parallel computation of Backward Step Flow
In this study, a parallelized CFD solver was developed in PETSc, using C programming language, to solve Navier-Stokes equations in curvilinear coordinates at various Reynold’s number. Artificial compressibility was incorporated in the continuity equations to solve the momentum equations by pseudo marching in time. The governing equations were discretized by employing a second-order accurate central difference scheme whereas a fourth-order RK4 scheme was used for time marching. The code used Message Passing Interface (MPI) parallelization over distributed memory nodes.
2. Hyper-parametric tuning via Machine Learning with OpenMP
In this project, parametric tuning was performed with supervised Machine Learning approach to determine accurate parameters for Gaussian Process Regression, thereby decreasing the mean square error between predicted and observed data. The code was developed in C programming language, and was parallelized with OpenMP over shared memory node.
3. Turbulent flow around rectangular cylinder
To investigate the turbulent statistics in flow around rectangular cylinders at high Reynold’s number and make a comparison with those from Direct Numerical Simulations (DNS), RANS-based turbulence models were tested in OpenFOAM, an open source CFD solver. This study also investigated the effect of modeling various scales of motion on the flow characteristics by employing Partially-averaged Navier-Stokes equations.
